Environmental dynamics shape human learning: change points versus random walks

Abstract

Adaptive learning requires interpreting prediction errors in light of environmental dynamics. However, environments may differ not only in when changes occur, but also in how they occur: some nonstationary processes drift gradually, whereas others change abruptly. This raises the question of whether humans can adapt their learning to the generative structure of their environment. Here, we compared learning in two canonical nonstationary environments: abrupt change points versus gradual random walks. Across two behavioral experiments using a predictive inference task and a unified Bayesian framework, we show that participants adjust their learning to the dynamics of their environment. Cross-validated model comparison showed that the environment-matched model best explained behavior in nearly all participants. Human apparent learning rates also showed the predicted dynamics-specific signature, increasing more sharply with prediction-error magnitude under change points compared to random walks. Participants tracked the changing latent variance (environmental stochasticity) online in both environments, jointly with the latent mean, through coordinated adjustments of paddle width and location. In the change-point environment, their width adjustments further distinguished changes in variance from changes in mean. Together, these results show that humans adapt their learning to both how latent states change and which properties are changing. They establish a unified computational account of adaptive learning across different environmental dynamics.